Design And Implementation Of Fact-checking System Based Deep Learning For Biomedical Text

Since the outbreak of COVID-19,a large number of false information related to COVID-19 has been widely spread on the Internet,which seriously threatens the health of people and the safety of the network environment,and also arouses social attention to false information in the biomedical domain.The purpose of fact-checking task is to verify the truth or falseness of a given claim based on evidence,and the research of fact-checking for biomedical text has far-reaching significance to curb the flood of false information in the biomedical domain.Aiming at building a fact-checking system for biomedical text,this thesis deeply analyzes the shortcomings of the existing fact-checking methods for biomedical text,and with the help of deep learning and natural language processing technologies,the following work is carried out on how to introduce semantic role relevance and the design and implementation of the system:(1)The fact-checking for biomedical text includes three sub-tasks: abstract retrieval,evidence extraction and fact verification.This thesis proposes a fact-checking model SRRM based on semantic role relevance modeling to solve the problem that the existing fact-checking methods for biomedical text ignore the rich and valuable semantic role level information in claim and evidence in fact verification stage.The model uses a semantic role heterogeneous graph attention network to model the semantic relevance of claim and sentences in abstract at the semantic role level.In addition,the SRRM model uses the framework of multi-task learning to jointly learn three sub-tasks in fact-checking.Experimental results on benchmark dataset show that the proposed method has better detection performance than the existing methods.The ablation experiment results demonstrate the importance of semantic role relevance modeling for fact-checking for biomedical text.(2)This thesis designs and implements a fact-checking system for biomedical text.In this thesis,the requirements of the system are analyzed in detail from functional and non-functional aspects,and the overall architecture,each module and database of the system are designed.The system consists of four modules: original data module,data processing module,model prediction module and result display module,and the detection results based on SRRM model are displayed on the front page in the form of text.This thesis makes a detailed test of the fact-checking system from the point of view of requirements,and the test results show the effectiveness of the fact-checking system for biomedical text of this thesis.